NovaFuse: Voice-First AI Study & Interview Agent

INSPIRATION Traditional learning and interview prep apps rely on text inputs, flashcards, or multiple-choice buttons. But real exams, defenses, and job interviews happen out loud under pressure. NovaFuse replaces silent reading with a voice-first conversational AI partner that challenges your knowledge through live spoken dialogues, probes your reasoning, and helps you master complex concepts. WHAT NOVAFUSE DOES NovaFuse is a real-time, voice-first study and mock interview platform featuring two specialized modes: Active Recall Revision Mode: Upload notes or syllabus outlines. The AI agent acts as a Socratic tutor, asking conceptual questions, following up on misunderstandings, and adapting difficulty based on your performance streak. STAR Mock Interview Mode: Select your target role and difficulty. The agent conducts a realistic interview, probing answers using the STAR methodology (Situation, Task, Action, Result) and evaluating technical depth. Adaptive Weak-Topic Memory Loop: NovaFuse scores topic mastery (strong, partial, weak) in real time. Past weak areas are automatically injected into future sessions, ensuring continuous spaced reinforcement. Post-Session Analytics: Receive instant takeaways, STAR rubric breakdowns, performance grades, and transparent per-session cost tracking ($0.075 per minute). HOW WE BUILT IT Audio Pipeline: Custom 24kHz mono PCM16 recording and playback worklets using the Web Audio API for sub-second latency and smooth barge-in interruptions. Voice Agent Integration: Connected via WebSocket to the AssemblyAI Voice Agent API for unified speech-to-text, LLM reasoning, neural semantic turn detection, and high-fidelity speech synthesis. Backend and Database: Node.js, Express, and WebSocket relays with PostgreSQL and Prisma ORM, backed by autonomous self-healing schema migrations and offline file storage fallback. Frontend: React 18, TypeScript, Tailwind CSS, and Vite.